another feature request

[Stephen Weeks]

> One hates to see the extra cost on a handle expression, but it can be
> very difficult to diagnose these uncaught exceptions.  A compile-time
> option might be suitable (with default to diagnose).

The cost is only when the handler runs, not when it is installed, so it
shouldn't be that bad.  Also, MLton's whole-program optimization might even get
rid of the string list ref for programs that don't use SMLofNJ.exnHistory.  In
any case, it's easy to add the compile-time bool that controls whether or not 
handlers muck with the field.  Although, I would probably make the default be to
not muck.

> I have just sent a mail message with an uncaught exception.  I hope it
> is in your code and not mine, but it's not entirely clear to me...

It's definitely a MLton bug (or bugs!).  I'll send more shortly.